Microsoft switches internal workloads from OpenAI to in-house MAI models
Microsoft is transitioning some internal workloads in Excel and Outlook from third-party AI models by OpenAI and Anthropic to its own MAI model family to reduce substantial token costs. Despite this move, Microsoft continues its partnership with OpenAI through 2032 as the industry shifts toward more cost-efficient in-house AI infrastructure.
Key Takeaways
- Internal Microsoft prompts for Excel and Outlook are being rerouted to the 35B-parameter MAI-Thinking 1 reasoning model.
- The 2032 expiration of the OpenAI partnership and non-exclusive licenses are driving internal model independence.
- Anthropic models were specifically identified as prohibitively expensive by Microsoft AI CEO Mustafa Suleyman.
- The MAI-Thinking 1 model achieves performance parity with Anthropic’s Claude Opus 4.6 on SWE Bench Pro coding tasks.
Why It Matters
This move signals the end of the uncurbed 'tokenmaxxing' era for enterprise tech giants. By swapping expensive third-party APIs for in-house 'good enough' models for routine tasks, Microsoft is establishing a tiering strategy that preserves premium models for complex reasoning while protecting margins on ubiquitous software. For the broader industry, this confirms that vertical integration of the AI stack is now a financial necessity, not just a technical preference. Watch for whether OpenAI responds by slashing API pricing for its 'O' series models to discourage further first-party substitution.
Additional Context
The transition away from premium third-party APIs is part of a broader industry crackdown on 'tokenmaxxing' activities that have strained corporate budgets. Per The Information (June 2026), Meta recently dismantled its 'Claudeonomics' leaderboard—which gamified AI usage—after employee token consumption neared a pace that would cost billions annually. Similarly, per ITPro (June 2026), Uber implemented strict monthly spending caps of $1,500 per employee for AI tools after exhausting its entire 2026 coding assistance budget in just four months.
Market dynamics are also being reshaped by aggressive pricing from open-source alternatives. Per Reuters (May 2026), China’s DeepSeek made its V4-Pro model pricing permanent at approximately 87 cents per million output tokens. This undercuts Anthropic’s premium offerings by nearly 20-fold, providing a benchmark for the 'cost-efficiency' targets Microsoft is likely chasing with its MAI family. Microsoft's drive for autonomy follows a significant April 2026 amendment to its OpenAI pact, which transitioned their partnership to a non-exclusive license through 2032 and removed Microsoft's revenue-share obligations to the AI lab.
Read full article at siliconangle.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source